Abstract: With the rapid development of economy and society,higher requirements are put forward for the speed and quality of information dissemination. In the field of computer vision,super-resolution reconstruction also make it easier for people to obtain high-resolution images. However,most current methods focus on improving the overall quality of images,and the processing results for specific targets contained in the image,especially small-sized objects,are not very satisfactory. This paper proposes new generative adversarial network,which uses real-ESRGAN as the baseline network framework,introduces Vision Transformer to enhance the self-attention mechanism and replaces the activation function. The feasibility of the algorithm is verified by designing a target detection ablation test on the algorithm for the targeted-designed dataset. The study results show that the model can generate higher quality images when facing images containing small objects,and it is superior to the original model in terms of mainstream super-resolution reconstruction effect indicators. 

Key words: super-resolution reconstruction,generative adversarial network,self-attention